VisRAG
VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents
Shi Yu, Chaoyue Tang, Bokai Xu, Junbo Cui, Junhao Ran, Yukun Yan, et al. · 2024-10-14
- Management
- T1 Static Memory
- Memory is built once and reused without change during deployment.
- Representation
- Source- and modality-specific
How experience becomes memory
Each path is one memory component, traced from what is stored to how it is found again.
visual document corpus
- Represented asRetained perceptual source evidenceSource- and modality-specific
- Stored inVector-indexed collection
- Retrieved bySimilarity search
- Source evidenceSource directly available
Document pages remain in image form; a VLM encodes each page for cosine-similarity retrieval.Full-paper method / memory representation and retrieval · PDF p. 5 · Figure 2
Retrieved page images, rather than their index embeddings, are supplied to the VLM generator.Full-paper method / memory representation and retrieval · PDF p. 4
Related systems
- UniversalRAGT1 · Video & streaming
- AtlasVAT1 · Image
- camroll-agentT1 · Image
- HiKEYT1 · Image
- MemShotT1 · Image
- MMAT1 · Image